Update on "[ET-VK] Adding batch processing in x axis to conv2d dw shader by caching input texel for reuse."

This diff adds batch processing in the x axis to the conv2d dw shader by reusing input texel overlapping between consecutive tiles. The changes include modifying the glsl code for the conv2d dw output tile, adding a new parameter to the yaml file, and modifying the Convolution.cpp file to use the new parameter.

Differential Revision: [D67868671](https://our.internmc.facebook.com/intern/diff/D67868671/)

[ghstack-poisoned]

Author: trivedivivek

.lintrunner.toml

@@ -321,6 +321,7 @@ command = [
     '--config=.mypy.ini',
     '--show-disable',
     '--',
+    '--explicit-package-bases',
     '@{{PATHSFILE}}'
 ]
 init_command = [

backends/cadence/aot/compiler.py

@@ -131,7 +131,10 @@ def quantize_pt2(
     Prepare, convert and fuse the model using the given quantizer.
     Returns a GraphModule with the quantized model.
     """
-    # Quantizer
+    # Make the model inference mode by calling model.eval()
+    model.eval()
+
+    # Instantiate the quantizer to CadenceQuantizer if not supplied
     if not quantizer:
         quantizer = CadenceQuantizer()
 

docs/source/conf.py

@@ -197,7 +197,7 @@
 
 # Custom directives defintions to create cards on main landing page
 
-from custom_directives import (
+from custom_directives import (  # type: ignore[import-not-found]
     CustomCardEnd,
     CustomCardItem,
     CustomCardStart,

examples/models/llama/export_llama_lib.py

@@ -79,7 +79,14 @@
 verbosity_setting = None
 
 
-EXECUTORCH_DEFINED_MODELS = ["stories110m", "llama2", "llama3", "llama3_1", "llama3_2"]
+EXECUTORCH_DEFINED_MODELS = [
+    "stories110m",
+    "llama2",
+    "llama3",
+    "llama3_1",
+    "llama3_2",
+    "static_llama",
+]
 TORCHTUNE_DEFINED_MODELS = ["llama3_2_vision"]